vbatts/llama.cpp
1
0
Fork 0
Code Issues Pull requests Projects Releases Packages Wiki Activity Actions

threadpool: update calling thread prio and affinity only at start/resume

Browse source 
This avoids extra syscalls for each graph_compute()
This commit is contained in:
Max Krasnyansky 2024-08-26 13:10:11 -07:00 • committed by fmz
parent 8186e9615f
commit 658f16c330
1 changed files with 14 additions and 6 deletions
Download patch file Download diff file Expand all files Collapse all files

20
ggml/src/ggml.c
View file

@ -19237,6 +19237,12 @@ static void ggml_graph_compute_kickoff(struct ggml_compute_threadpool * threadpo
atomic_fetch_add_explicit(&threadpool->n_graph, 1, memory_order_relaxed);
if (threadpool->pause) {
// Update main thread prio and affinity to match the threadpool settings
ggml_thread_apply_thread_priority(threadpool->prio);
if (ggml_thread_cpumask_is_valid(threadpool->workers[0].cpumask)) {
ggml_thread_apply_affinity(threadpool->workers[0].cpumask);
}
// resume does cond broadcast
ggml_resume_threadpool_locked(threadpool);
} else {
@ -19324,6 +19330,14 @@ static struct ggml_compute_threadpool * ggml_create_threadpool_impl(
}
ggml_thread_cpumask_next(tpp->cpumask, workers[0].cpumask, tpp->strict_cpu, &cpumask_iter);
if (!threadpool->pause) {
// Update main thread prio and affinity at the start, otherwise we'll do it in resume
ggml_thread_apply_thread_priority(threadpool->prio);
if (ggml_thread_cpumask_is_valid(threadpool->workers[0].cpumask)) {
ggml_thread_apply_affinity(threadpool->workers[0].cpumask);
}
}
#endif // GGML_USE_OPENMP
return threadpool;
@ -19380,12 +19394,6 @@ enum ggml_status ggml_graph_compute(struct ggml_cgraph * cgraph, struct ggml_cpl
ggml_graph_compute_thread(&threadpool->workers[0]);
}
#else
// Update main thread prio and affinity to match the current threadpool
ggml_thread_apply_thread_priority(threadpool->prio);
if (ggml_thread_cpumask_is_valid(threadpool->workers[0].cpumask)) {
ggml_thread_apply_affinity(threadpool->workers[0].cpumask);
}
// Kick all threads to start the new graph
ggml_graph_compute_kickoff(threadpool);